‘Not even close to rat’: Facebook AI director believes Terminator scenario unlikely in near future
“We’re very far from having machines that can learn the most basic things about the world in the way humans and animals can do,” Yann LeCun said in an interview with the Verge.
Although artificial intelligence has “superhuman performance” in particular areas, “in terms of general intelligence we’re not even close to a rat,” he stated.
While popular culture fuels our fears of computers gaining self-awareness, LeCun emphasized there was “no danger in the immediate or even medium term” of unleashing Skynet. “There are real dangers in the department of AI, real risks, but they’re not Terminator scenarios.”
Despite the breakthroughs in the AI field, such as self-driving cars, software reading and interpreting medical images or Google’s AI defeating world’s master at Go, these are examples of “very narrow intelligences.”
Developed by Google’s AI subsidiary Deep Mind, the program managed to beat the top Go player because it used “reinforcement learning.” The AI basically played “millions of games over the course of a few days or few weeks, which is possibly more than humanity has played at a master level since Go was invented thousands of years ago,” LeCun explained.
To develop a general intelligence, the machines need to learn through building their own internal models of the world, LeCun said, adding that Facebook is interested in getting AI to explore the world by watching movies or reading books.
'General fearfulness doesn’t help, real danger would come from self-improving seed' - AI pioneer on robot takeover https://t.co/fbTztHabp8— RT (@RT_com) March 27, 2017
LeCun believes that the next “big thing” will be virtual assistants that aren’t scripted and “frustrating” to talk with, but truly useful. “But we’re not going to get that unless we can find some way of getting machines to learn how the world works by observation,” he said.